Research On DSP And Super-resolution Based On OCD Platform Acceleration

With the development of technology and manufacture of semiconductor industry, the demand to the critical dimension of semiconductor component has become higher, it put forward higher requirements for correct detection of critical dimension(CD) in process of manufacture. Traditional methods of measurement of CD including scanning electron microscopy(SEM), atomic force microscopy(AFM), scanning probe microscope(SPM). etc. The advantage of these methods have visible, but suffer from the shortage such as: the measuring environment, precision, efficiency, and may not be non-destructive for all kinds of samples. These prevent them from the application of on-line measurement, which is important in manufacture. On the contrary, the optical critical dimension(OCD) measurement with lower cost, higher accuracy, faster, and no damage to all kinds of samples.OCD software platform mainly includes optimizing module and simulation module. The optimizing module takes charge of modeling the sample, the simulation module responsible for computing and analyzing the sample model output optical response, and comparing to the real optical information, and then correct the old sample module, repeat above steps until the output optical response is same to experimental results, the new sample’s dimension is the material’s dimension.According to optical diffraction theory, rigorous coupled wave analyses(RCWA) are the most commonly used method to calculate sample’s optical response for the simulation module, and the performance of OCD software is depend on the performance of simulation module. The essence of RCWA is to solve the Maxwell equations, normally, it perform a large number of matrix computing(specific calculate amount is depend on levels of diffraction) and takes a lot of time, especially eigenvalue and eigenvector, and it limits the efficiency of simulation module.In this work, we try to use TI’s TMS320C6678 to accelerate the eigenvalue and eigenvector of the matrix. Digital Signal Processor(DSP) is usually not suitable to general computing, but with the development of DSP technology, its ability of multiply and memory access performance has been promoted, and they are very helpful to the matrix operation performance. Another reason why we use 6678 is that its easy extension and low power consumption.With the smaller dimension of micro-electronic components, another method to increase accuracy of measurement is super-resolution technology. The super-resolution can break the limits of component’s diffraction and achieve high resolution. The last chapter of this work introduces super resolution technology and try to use it in our OCD system.